Facial recognition has become an essential technology in sophisticated surveillance systems, tackling significant issues related to public safety, access control, and anomaly detection. The challenges posed by surveillance settings, which include varying lighting conditions, occlusions, and low-resolution video streams, require solutions that are highly versatile and resilient. This study presents a hybrid face recognition framework that combines the feature extraction power of the pretrained VGG19 network with the adaptability of convolutional neural networks (CNNs), specifically tailored for dynamic surveillance situations. The design utilizes VGG19’s hierarchical feature extraction while incorporating custom CNN layers to strengthen its resilience against pose changes, low-light environments, and partial occlusion. Notable advancements consist of sophisticated data augmentation methods, transfer learning approaches, and model compression techniques aimed at enhancing scalability and efficiency in deployment. Experimental tests indicate that the suggested hybrid model reaches outstanding accuracy, exceeding 95%, and maintains real-time processing capacity. This method highlights the ground-breaking possibilities of hybrid deep learning frameworks, providing a scalable, efficient, and trustworthy solution for contemporary surveillance systems, ranging from localized security measures to extensive monitoring networks.